In December 2025 I counted the words in five years of this newsletter. I wrote about coffee 54 times. This is the ledger: one row per study, oldest at the bottom, with the caveats I attached at the time and links to the issue and the deep dive.
The second table was added in September 2026, and it is the more honest half of the page. Building the first table made it obvious that I have been reading coffee research the way most people read the news, which is to say whatever came out this month. The umbrella review that everyone cites, the cohort of 468,629 people, the meta analysis behind the dementia number I have quoted twice, the entire literature on coffee and Parkinson's disease: none of it has ever appeared here. So it is listed below, unread and admitted to.
Some ground rules for reading both tables. Nearly everything here is observational. Coffee drinkers are not a random sample of humanity, and healthy user bias runs through the whole column. The industry group ISIC (Illy, Lavazza, Peet's, and others) funds a surprising share of the mechanistic work. And I have declared my bias in the newsletter more times than I have declared it to my accountant. The ledger exists partly so the bias has to look at itself in a table.
| Date | Journal / source | What it found, and the caveat I attached | Links |
|---|---|---|---|
| 2026-09-04 | npj Science of Food (Cao, An, Du, Xu, Wang, Lu; Peking University). Published Dec 2025, corrected Feb 2026. | Mendelian randomization on UK Biobank and MiBioGen: unsweetened filter coffee tracks with more Veillonella, which tracks with lower HbA1c, with 43.33% of the effect running through that genus. The finding that matters is the convergence: the Cork feeding study, four months later and sharing no data or authors, found Veillonella blooming when coffee was reintroduced. Every estimate in the mediation chain sits close to zero, and microbiome genetic instruments are weak. The paper was corrected because the results and tables had decaf and caffeinated labelled backwards. Mendelian randomizationPreparation, not caffeinePublished correction |
Deep diveIssue: pending |
| 2026-09-04 | Nature Communications (Boscaini, Bastiaanssen, Moloney, Cryan; APC Microbiome Ireland). ISIC funded. | 62 healthy Irish adults, 31 habitual drinkers and 31 never drinkers. Two week caffeine washout, then three weeks of blinded caffeinated or decaf instant. Habitual drinkers carry a distinct set of gut strains and metabolites that drift back toward the nondrinker profile off coffee; diversity does not change. Mood questionnaires improved on either coffee; cortisol did not move. Memory improved only in the decaf group (n = 15), which the authors flag as a possible practice effect. The viral "coffee drinkers have worse memory" headline is a secondary outcome the authors say the study was not powered for. Citation audit, September 2026: 542,000 accesses and 142 news outlets, but six citations in four and a half months, none flagged influential, four of five in MDPI journals, and no attempted replication. Decaf shares the benefitRCT (partial) |
Deep diveIssue: pending |
| 2026-08-17 | Multi source synthesis (ergogenic aids literature) | A Division 1 rower's coach prescribed caffeine, sodium bicarbonate, and beta alanine. Caffeine dosed around races has the best evidence of the three. My instinct that this was nonsense was wrong; what is unsafe in patients with multiorgan disease can be reasonable in an athlete with none. Caffeine as performance aid |
Deep diveIssue 08-17-26 |
| 2026-08-03 | Circulation, AHA Scientific Statement (Marcus, Hu, et al.) | The American Heart Association's first formal position on caffeine. Up to about 400 mg a day (three to five cups) is safe; moderate intake tracks with lower coronary disease, stroke, heart failure, atrial fibrillation, hypertension, and type 2 diabetes, in a J curve bottoming near two to four cups. Decaf shares the diabetes benefit, so that one is not caffeine. Unfiltered coffee raises LDL via cafestol; a paper filter fixes it. Energy drinks and caffeine powders are a different, riskier category. Almost all observational. Bias disclosure: we upgraded the espresso machine that week. Decaf shares the T2D benefit |
Deep diveIssue 08-03-26 |
| 2026-06-29 | Cremieux blog + OpenEvidence literature review | A blog on association and causation fallacies, including the early 1980s literature asking whether hot drinks cause cancer (they do not; older people who drink hot drinks get more cancer). The coffee review was comforting: coffee is safe, caffeine is mostly safe, and coffee is associated with lower mortality, cardiovascular disease, type 2 diabetes, depression, dementia, liver fibrosis, several cancers, and chronic kidney disease. Review |
Issue 06-29-26 |
| 2026-02-16 | JAMA (Nurses' Health Study + Health Professionals Follow up Study; 131,000 people, 43 years) | Two to three cups of caffeinated coffee, or one to two of caffeinated tea, a day was associated with an 18% lower risk of dementia. Decaf showed no benefit. More than two to three cups added nothing, and the association held regardless of genetic risk. The issue where the corner got its formal name: Adam's Coffee Confirmation Bias Corner. Decaf: no benefit |
Issue 02-16-26 |
| 2025-12-01 | BMJ Mental Health (436 Norwegians with severe mental illness) + Frontiers fMRI study (ISIC funded) | Three to four cups a day was associated with the longest telomeres, roughly five years younger biological age after adjustment; why the cohort was people with severe mental illness is not well explained. Separately, fMRI 30 minutes after coffee quieted mind wandering regions and boosted visual and decision making areas; caffeine dissolved in hot water did only the first. Something about coffee beyond caffeine, or the ritual, seems to matter. This is where I first learned that ISIC funds coffee research, and added a filter. Cross sectionalISIC first flagged |
Issue 12-01-25 |
| 2025-11-17 | JAMA (DECAF trial, 200 patients) + a review of additives | The antithesis of what most of us were taught: in a randomized trial of 200 patients with atrial fibrillation or flutter, recurrence was lower in the group told to drink caffeinated coffee than in the group told to abstain. "We shall see if these data hold up." (The AHA statement nine months later leaned on it.) Also: the mortality benefit in the cohort literature belongs to black coffee; sugar and saturated fat in the cup erase it. RandomizedCaffeine specific |
Issue 11-17-25 |
| 2025-11-10 | Nutrition Journal (Li et al., Inner Mongolia Medical University; NHANES older adults) | Self reported high volume coffee intake (480 g a day or more) was associated with 42% lower odds of poor cognitive performance on memory and processing speed. Decaf did not show the same benefit. Not a typical research institution, and a cross sectional NHANES analysis. Cited in September 2026 by the Cork microbiome paper as one of its motivating studies. Decaf: no benefit |
Issue 11-10-25 |
| 2025-10-06 | 2022 review of chlorogenic acid intake (PubMed) + X thread on roasting | Chlorogenic acids, the main coffee phenolics, are reduced by darker roasting. Meta analyses of randomized trials put daily intakes of 13.5 to 1200 mg against lower fasting glucose, better glucose tolerance, weight management, and lower blood pressure in hypertensives, with a dose response above 200 mg. I do not advocate CGA for medicinal purposes. I mostly want to feed my confirmation bias. Mechanism |
Issue 10-06-25 |
| 2025-09-08 | Nutrients (Hamburg City Health Study, 2,316 participants, MRI) | Three to four cups a day was associated with better MRI markers of vascular damage, neurodegeneration, and microstructural integrity than under one cup. Cross sectional, self reported, one homogeneous German population, coffee type unknown. Comforting, imperfect. Cross sectional |
Issue 09-08-25 |
| 2025-06-09 | Nurses' Health Study (47,000 women, 30 years; presented at Nutrition 2025) | Higher caffeine intake, mostly from coffee, was associated with a higher likelihood of healthy aging: reaching 70 free of eleven major chronic diseases with intact physical function, mental health, and cognition. Decaf and tea showed no association. Soda, another caffeine source, was associated with a 20 to 26% lower likelihood of healthy aging. Not all caffeine is created equal. Decaf: no benefit |
Issue 06-09-25 |
| 2025-02-17 | Cool Infographics (caffeine content by brand) | A pause in the parade: caffeine can be inappropriate for people with specific conditions, and the caffeine content of commercial coffee varies far more by brand than most drinkers assume. Caveat entry |
Issue 02-17-25 |
| 2025-01-13 and 2025-02-10 | European Heart Journal (NHANES, about 40,000 adults, median 9.8 years) | When you drink it matters. People who drank two to three cups mostly in the morning had lower cardiovascular and all cause mortality than nondrinkers; all day drinkers showed no benefit, plausibly through sleep disruption. Observational, self reported timing. Covered twice, then a third time in the January 2026 health habits roundup, which tells you how much I liked it. Observational |
Issue 01-13-25Issue 02-10-25 |
| 2025-01-06 | Nature Microbiology (Manghi et al., 2024; 22,000+ gut microbiome samples) | Across multiple cohorts, coffee consumption was the dietary factor most strongly and consistently tied to gut microbiome composition, through one bacterium, Lawsonibacter asaccharolyticus. Coffee directly stimulated its growth in culture, and the organism appears to process coffee compounds such as quinic acid. One of the most direct links between a specific food and a specific microbe. The 2026 Cork study is the interventional follow on to this line of work and cites it. Mechanism |
Issue 01-06-25 |
| 2023-11-27 | Cell & Bioscience (Taiwan) | Population databases associated coffee consumption with lower COVID risk, and bench experiments showed coffee (ground or instant, regular or decaf, with or without milk and sugar) inhibited coronavirus entry into cells in a dish. The authors proposed "a coffee drinking plan" for the post COVID era. Early and a bit bold. Starbucks, as best I could tell, did not fund it. Decaf shares the effect (in vitro) |
Issue 11-27-23 |
| 2020-09-15 | Meta analyses via PubMed (Nephrology Nurses Week roundup) | The first entry, and the nephrology one. Coffee consumption is associated with a lower risk of chronic kidney disease and end stage kidney disease. Takeaway for otherwise healthy adults: low sodium, water not soda (diet or regular), and "drink coffee, maybe 2+ cups a day," with the usual advice to get individualized guidance rather than take generic advice from a random kidney doctor who emails you every day. Meta analysis |
Issue 09-15-20 |
| 2026-08-04 | VoxEU / CEPR (Voth & Hersh, 2009). Not a health study. | Footnote row. An economics column arguing that real wage indices miss the welfare gain from new goods, with coffee as the worked example. On the Cutting Room Floor and currently suppressed pending sign off on a WAiR added section. Listed here only so the ledger is complete. Economics, not health |
Deep dive (suppressed) |
| Paper or question | Citation | Why it belongs here, and why its absence is awkward |
|---|---|---|
| The umbrella review | Poole R, et al. BMJ 2017;359:j5024 | An umbrella review of meta analyses across every health outcome anyone has studied coffee against. It is the most cited coffee safety document in existence and the actual source of the "three to four cups is the sweet spot" claim that appears in four rows of the table above. I have repeated its conclusion for years without ever reading or citing it here. Foundational |
| The other umbrella review | Grosso G, Godos J, Galvano F, Giovannucci EL. Annu Rev Nutr 2017;37:131 | The companion synthesis, and the one nutrition researchers tend to cite over the BMJ paper. Between the two, most of what I believe about coffee is downstream of work I have not covered. Foundational |
| The big mortality cohort | Simon J, et al. Eur J Prev Cardiol 2022;29:982 | 468,629 people in the UK Biobank without cardiovascular disease at baseline, with light to moderate coffee tracking to lower all cause mortality, cardiovascular mortality, and stroke. The archive is full of UK Biobank studies. None of them is about coffee. Foundational |
| The dementia meta analysis | Liu QP, et al. Nutrition 2016;32:628 | The prospective cohort meta analysis behind the figure that coffee is associated with a 27% lower incidence of Alzheimer's disease. That number is the first thing the Cork paper reaches for in its introduction, and I have gestured at it more than once without ever going to the source. Foundational |
| The depression meta analysis | Wang L, Shen X, Wu Y, Zhang D. Aust N Z J Psychiatry 2016;50:228 | Coffee and caffeine against depression risk across observational studies. It has appeared in this newsletter exactly once, as a single word inside a list someone else wrote. Foundational |
| Parkinson's disease | Noyce AJ, et al. Ann Neurol 2012;72:893 and Qi H, Li S. Geriatr Gerontol Int 2014;14:430 | The inverse association between coffee and Parkinson's disease is one of the most consistent findings in nutritional neurology, with a clean dose response. In six years and 548 documents, the only times Parkinson's appears here are deep brain stimulation and Neuralink. This is the largest single hole in the corner. Foundational |
| The microbiome origin study | Asnicar F, et al. Nat Med 2021;27:321 (PREDICT 1) | 1,098 deeply phenotyped individuals, more than 150 dietary components tested, and coffee came out as the food most strongly correlated with gut microbiome composition. The January 2025 row covers the follow on study that identified Lawsonibacter. This is the paper that made anyone look. Foundational |
| Caffeine sensitivity genetics | Rétey JV, et al. Clin Pharmacol Ther 2007;81:692 and Landolt HP. Sleep 2012;35:899 | Variation in the adenosine A2A receptor gene (ADORA2A) explains a good deal of why the same cup wrecks one person's sleep and does nothing to another's. The Cork study genotyped its participants for exactly these variants. Nothing on this in the archive. Foundational |
| CYP1A2, appetite and BMI | Gkouskou KG, et al. Int J Obes 2022 (online Sept 2021) | A crossover trial of 18 plus a cohort of 284, stratified by the rs762551 variant that sorts people into rapid, intermediate and slow caffeine metabolizers. Coffee lowered energy intake, fat intake and appetite, and tracked with lower BMI, only in rapid metabolizers. Small and borderline, but it is the other half of the genetics thread and it adds an outcome the table above does not touch. Surfaced by Nature's related article list alongside the Cork study. Candidate |
| Coffee and athletic performance, minus the caffeine | Grzelczyk J, et al. Nutrients 2026;18:2690 | A narrative review that cites the Cork study and reaches the opposite emphasis: coffee before exercise supports endurance and blunts perceived exertion and muscle pain, effects on strength and power are inconsistent, and evidence for the non caffeine compounds is "limited and largely indirect, indicating a modulatory rather than ergogenic role." A useful corrective to hold against the decaf enthusiasm above, and a direct follow up to the ergogenic aids row. Candidate |
| The clinical questions I have never asked | No paper selected | Three gaps with zero hits across the whole archive. Coffee and gout or urate, and coffee and kidney stones, which is conspicuous given that the oldest row in this ledger is the chronic kidney disease one and I do this for a living. And coffee in pregnancy, which is the one common clinical situation where the cheerful story genuinely reverses and the advice becomes a limit rather than a permission. Open |
Read down the first column and one question keeps splitting the literature: does decaf get the benefit? The answer sorts the outcomes surprisingly cleanly. Where the benefit is metabolic, microbial, or anti inflammatory, decaf tends to share it. Where the benefit is cognitive or cardiac rhythm, it tends to belong to caffeine.
Decaf shares the benefit
- Type 2 diabetes risk (AHA statement, 2026)
- Gut microbiome and metabolite shifts, mood questionnaires, sleep (Cork, 2026)
- Anti inflammatory cytokine response (Cork, 2026)
- Coronavirus entry in vitro (Taiwan, 2023)
Caffeine specific, decaf shows nothing
- Dementia risk (JAMA, 131,000 people, 2026)
- Cognitive performance in older adults (NHANES, 2025)
- Healthy aging (Nurses' Health Study, 2025)
- Atrial fibrillation recurrence (DECAF trial, 2025)
- Trait anxiety and attention (Cork, 2026)
- Endurance and perceived exertion (Nutrients review, 2026)
A third axis, added September 2026: preparation. The HbA1c study does not sort onto either side, because its claim is about how the coffee was made rather than what was taken out of it. Unsweetened and filtered now has three independent arguments behind it: a paper filter removes the cafestol that raises LDL, the mortality signal in the cohort data belongs to coffee without much sugar or fat in it, and the glycemic effect in the Mendelian randomization data attaches specifically to unsweetened filter coffee. Worth watching whether preparation turns out to explain some of what gets attributed to caffeine.
Recurring caveats, so I do not have to keep retyping them. Healthy user bias (moderate coffee drinkers are healthier before the first cup). Self reported intake. Coffee versus caffeine is nearly impossible to separate in cohort data, and at least one journal has now published the two labelled backwards. The J curve means "more" stops helping around four cups. Additives matter. Unfiltered coffee raises LDL. Energy drinks and caffeine powders are not coffee and the data on them point the other way. And ISIC money is in a fair share of the mechanistic studies.
Six years, eighteen entries, two randomized trials, and no study of moderate coffee that found net harm. That is either the most durable confirmation bias in this newsletter or a fairly consistent literature. It is probably both. The second table is the part I would rather not have written: the umbrella review everyone cites, the cohort of 468,629, and the entire Parkinson's literature have never appeared here, which means I have been keeping a careful tally of the news and calling it a body of knowledge. The advice has not changed since the first row in 2020. Two to four cups, mostly in the morning, filtered, without much in it.
Sources
Each tier one row links to the WAiR issue where the study was covered and, where one exists, to the WAiR deep dive. Primary citations for each study live in those issues and artifacts. Newest entries: Cao Z, et al. npj Sci Food (2025), doi.org/10.1038/s41538-025-00655-w, with its Author Correction; and Boscaini S, et al. Nature Communications 17:3439 (2026), doi.org/10.1038/s41467-026-71264-8.
Maintenance note: tier one was assembled on 4 September 2026 from a full text search of the wair.ajwein.com archive for "coffee," "caffeine," and "decaf," restricted to entries where a study, review, or dataset was discussed (mentions of "coffee read of the day" and AI art prompts excluded). Tier two was assembled the same day by checking the 97 references of the Cork microbiome study, plus Nature's related article list for it, against the full text of all 548 documents in that archive, and keeping only items with zero prior mentions. Citation and attention counts for the Cork study come from its Nature metrics page and a Semantic Scholar citation query, both on 4 September 2026.